Jobbench
JobBench: realistic white-collar professional tasks (65 main + 63 easy across 35 occupations). Agents work in a sandbox over heterogeneous reference files (CSV, xlsx, PDF, docx, SQLite) and submit file deliverables, graded by the official per-rubric pass/fail judge (weighted r…
- Domain
- rl-env
- License
- unknown
- Published
- Jul 2026
Cite
Notes
Only stored in your browser.
Top score 52.9 by Kimi K3 - 1 model reporting
Top models
1
FAQ
- What is Jobbench?
- JobBench: realistic white-collar professional tasks (65 main + 63 easy across 35 occupations). Agents work in a sandbox over heterogeneous reference files (CSV, xlsx, PDF, docx, SQLite) and submit file deliverables, graded by the official per-rubric pass/fail judge (weighted r…
- What is the current top score on Jobbench?
- The top reported score is 52.9 by Kimi K3, across 1 model reporting.
- What license is Jobbench under?
- Jobbench is available under unknown.